Spaces:
Sleeping
Sleeping
File size: 1,878 Bytes
d20d20b c2af308 d20d20b d899b2a c51a031 d899b2a d20d20b b7ebcbb 15d43aa f2886e5 c51a031 d20d20b d899b2a c51a031 15d43aa 726fee7 d20d20b d899b2a c51a031 d20d20b c51a031 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 |
from fastapi import FastAPI
from pydantic import BaseModel
import requests
from llama_cpp import Llama
llms = {
"TinyLLama 1b 4_K_M 2048": {
"nctx": 2048,
"file": "tinyllama-1.1b-chat-v0.3.Q4_K_M.gguf",
"prefix": "### Human:",
"suffix": "### Assistant:"
},
"TinyLLama 1b OpenOrca 4_K_M 2048": {
"nctx": 2048,
"file": "tinyllama-1.1b-1t-openorca.Q4_K_M.gguf",
"prefix": "<|im_start|>system You are a helpfull assistant<|im_end|><|im_start|>user",
"suffix": "<|im_end|><|im_start|>assistant"
},
"OpenLLama 3b 4_K_M 196k": {
"nctx": 80000,
"file": "open-llama-3b-v2-wizard-evol-instuct-v2-196k.Q4_K_M.gguf",
"prefix": "### HUMAN:",
"suffix": "### RESPONSE:"
},
"Phi-2 2.7b 4_K_M 2048": {
"nctx": 2048,
"file": "phi-2.Q4_K_M.gguf",
"prefix": "Instruct:",
"suffix": "Output:"
},
"Mixtral MOE 7bx2 4_K_M 32K": {
"nctx": 32000,
"file": "mixtral_7bx2_moe.Q4_K_M.gguf",
"prefix": "",
"suffix": ""
},
"Stable Zephyr 3b 4_K_M 4096": {
"nctx": 4096,
"file": "stablelm-zephyr-3b.Q4_K_M.gguf",
"prefix": "<|user|>",
"suffix": "<|endoftext|><|assistant|>"
}
}
#Pydantic object
class validation(BaseModel):
prompt: str
llm: str
max_tokens: int = 512
nctx: int = 2048
#Fast API
app = FastAPI()
@app.post("/llm_on_cpu")
async def stream(item: validation):
model = llms[item.llm]
prefix=model['prefix']
suffix=model['suffix']
nctx = item.nctx if item.nctx is not None else model['nctx']
max_tokens = item.max_tokens if item.max_tokens is not None else 512
user="""
{prompt}"""
model = Llama(model_path="./"+model['file'], n_ctx=model['nctx'], verbose=False, n_threads=8)
prompt = f"{prefix}{user.replace('{prompt}', item.prompt)}{suffix}"
return llm(prompt, max_tokens=max_tokens) |